Research map / Label-efficient and robust learning / OOD and anomaly detection
RareCLIP: Rarity-aware Online Zero-shot Industrial Anomaly Detection
Jianfang He, Min Cao, Silong Peng, Qiong Xie
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Sequential Coreset Sampling instead of random sampling (ablation) | RareCLIP · MVTec AD | I-AUC: 88.13 -> 98.19 | Improved |
| Sequential Coreset Sampling instead of random sampling (ablation) | RareCLIP · VisA | I-AUC: 75.38 -> 94.40 | Improved |
| Sequential Coreset Sampling instead of greedy coreset (ablation) | RareCLIP · VisA | P-AUC: 88.46 -> 98.80 | Improved |
| K-means clustering sampling instead of random (ablation) | RareCLIP · MVTec AD | I-AUC: 88.13 -> 88.05 | No clear effect |
| add Loose Similarity (Y=1%) (ablation) | RareCLIP · MVTec AD | I-AUC: 97.00 -> 98.19 | Improved |
| add Loose Similarity (Y=1%) (ablation) | RareCLIP · VisA | I-AUC: 93.24 -> 94.40 | Improved |
| rarity threshold X=30% vs 5% (ablation) | RareCLIP · MVTec AD | I-AUC: 97.24 -> 98.19 | Improved |
| rarity threshold X=60% vs 30% (ablation) | RareCLIP · VisA | I-AUC: 94.40 -> 93.62 | Hurt |
| online memory update vs offline mode | RareCLIP · MVTec AD | I-AUC: 91.5 -> 98.2 | Improved |
| online memory update vs offline mode | RareCLIP · VisA | I-AUC: 86.1 -> 94.4 | Improved |
| indirect rarity branch with sequential sampling vs direct (RareCLIP-d) | RareCLIP · VisA | I-AUC: 93.5 -> 94.4 | Improved |
| indirect rarity branch vs direct (RareCLIP-d) | RareCLIP · VisA | P-F1-max: 48.1 -> 50.9 | Mixed |
Closest papers by meaning
- WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentation (CVPR 2023 · similarity 0.91)
- AA-CLIP: Enhancing Zero-Shot Anomaly Detection via Anomaly-Aware CLIP (CVPR 2025 · similarity 0.90)
- MoECLIP: Patch-Specialized Experts for Zero-shot Anomaly Detection (CVPR 2026 · similarity 0.89)
- AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection (ICLR 2024 · similarity 0.89)
- MuSc: Zero-Shot Industrial Anomaly Classification and Segmentation with Mutual Scoring of the Unlabeled Images (ICLR 2024 · similarity 0.88)
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
